Large language models (here GPT‑5) can originate nontrivial theoretical research ideas and contribute to derivations that survive peer review, if integrated into structured 'generator–verifier' human–AI workflows. This produces a new research model where models are active idea‑generators rather than passive tools.
— This could force changes in authorship norms, peer‑review standards, research‑integrity rules, training‑data provenance requirements, and funding/ethics oversight across science and universities.
Steve Hsu
2025.12.02
100% relevant
Steve Hsu reports a Physics Letters B paper (arXiv:2511.15935) whose central idea 'originated de novo from GPT‑5' and documents use of a generator–verifier pipeline with GPT‑5, Gemini and Qwen‑Max.
← Back to All Ideas